Seven wrist and 6 metacarpalphalangeal (MCP) regions from 5 subjects with arthritis were used to develop and validate 3D image acquisition and processing techniques. HDI values from 18 wrist and 9 MCP regions were obtained from 17 patients with active arthritis and compared with data from 10 wrist and MCP regions from 5 controls. The subjects included pediatric patients recruited from a single pediatric rheumatology practice and adult patients recruited from an academic rheumatology center. Diagnosis and classification of RA or JIA were made based on accepted ACR criteria or International League Against Rheumatism criteria 78. Active arthritis was defined as the presence of swelling and tenderness. The study protocol was approved by the University of Pittsburgh Institutional Review Board. All patients signed informed consent forms prior to inclusion in the study.

The image acquisition and processing technique is outlined in Figure 1. A forearm-based hand splint was designed to minimize movement and standardize hand position and pose between sessions (Figure 1a). Fixed objects, necessary to create and align 3D models from different sessions, were attached to the base of this splint. Scans were acquired using a Minolta Vivid 910 (Konica Minolta Sensing Americas, Inc., Ramsey, NJ, USA), a laser line triangulation scanner that produces a 640 × 480-pixel 3D image. Its manufacturer-reported resolution and accuracy are less than 0.2 mm in all axes. The camera was operated via a laptop computer using Polygon Editing Tool (version 1.22; Konica Minolta Sensing Americas, Inc.). Scans were acquired under the following standard conditions: camera positioned perpendicular to the subject's forearm at a stand-off distance of 0.8 m, camera height of 0.8 m, and camera declined to 45° from horizontal. Ambient room lighting was used during image acquisition. Two scans of each hand/wrist were acquired, one from the medial and the other from the lateral side. Rapidform2006 software (INUS Technology, Inc., Seoul, South Korea) was used to merge these two scans into a single, complete 3D model (Figure 1b). To follow patients longitudinally, it was essential to prevent minor wrist or hand rotation from session to session which might cause false variations in volume measurements. The fixation device constructed for this study prevented most such rotation. Small positioning changes that did occur were readily overcome by aligning the forearm and hand of models created across sessions, using the Rapidform co-registration function. Subsequent 3D models were constructed in the same fashion and then aligned to the reference model using the fiduciary markers and stable anatomic landmarks.

After model creation, two distinct computer-generated regions of interest (ROIs) were defined, one for the wrist and one for the 2nd-5th MCPs (green boxes in Figure 1b). The 2nd-5th MCP region was treated as a single ROI because the MCP joints are in juxtaposition to each other, and, in the case in which an MCP is swollen, it is impossible to determine where one MCP region ends and the adjacent one begins. The center of the wrist ROI was defined as the midpoint of the distance between the radial and ulnar styloids, whereas the center of the 2nd-5th MCP ROI was defined as the midpoint of the distance between the peaks of the 3rd and 4th MCP. Through trial and error, we determined that wrist ROI box dimensions of 9 cm in the medial-lateral plane, 4 cm in the proximal-distal plane, and 4 cm in the vertical plane and MCP ROI box dimensions of 10 cm in the medial-lateral plane, 2 cm in the proximal-distal plane, and 2 cm in the vertical plane encompassed maximal relevant data. These ROI boxes were created at the initial imaging session and remained fixed for all subsequent sessions. The wrist or MCP ROI was then extracted by deleting all data outside of the ROI box (Figure 1c). Volumes within the ROIs were then calculated. In addition, all points on the joint surface could be represented as distances in millimeters from the bottom plane of the ROI box. These distances could then be depicted as a color map (Figure 1d). We have established a surface distribution index (SDI), defined as one standard deviation (SD) from the mean of the all surface points-to-bottom plane distances. The SDI is a reflection of the surface shape, and distortions due to swelling will result in a change in SDI. The SDI data were generated using the 'Shell-Surface deviation' function in the Rapidform software.

Thermal data were acquired using a Meditherm medPro2000 thermoelectrically cooled microbolometer (Meditherm, Inc., Beaufort, NC, USA) and WinTES Thermal Evaluation Software (Compix, Lake Oswego, OR, Queensland, Australia). Unlike other commercially available thermal imagers, this sensor is specifically designed to measure temperatures found in the human body (10°C to 40°C). The device has a manufacturer-reported sensitivity and accuracy of less than 0.1°C and self-calibrates to an internal source at each pixel, avoiding the need for an external calibration target. Following International Academy of Clinical Thermology guidelines 9, subjects were asked to remove all jewelry and clothing covering the joints of interest and were given a 15-minute acclimation period prior to thermal imaging. All thermal images were obtained with the camera positioned directly over the hands. Ambient room temperature was 22°C ± 0.5°C. Skin emissivity was fixed at 0.98 1011. Thermal data were processed using specially designed code in Matlab (The MathWorks, Inc., Natick, MA, USA). With this code, centers for standard ROI boxes were selected (Figure 1e). The midpoint of the wrist or the midpoint between the 3rd and 4th MCPs served as the center of the thermal ROI boxes. A heat distribution index (HDI) was defined as twice the SD of all temperatures within the ROI 8. Relative frequency distributions were generated by plotting the frequency of temperatures in 1°C increments.

Wrist and MCP volume and shape vary across individuals. Therefore, pooled SDs were used to represent the overall measurement error SD when measuring volume and shape on multiple individuals. Excel XP (Microsoft Corporation, Redmond, WA, USA) and SAS 9.1 (SAS Institute Inc., Cary, NC, USA) were used for analysis. The average CV was used as a measure of overall CV. The ICC 13 was used as a measure of reliability 12. When comparing HDIs, group means were used to examine for significant differences using Student t tests. P values of less than 0.05 were considered significant. The area under the receiver operating characteristic (ROC) curve was used to assess overall sensitivity and specificity of thermal imaging 13.

We tested the reliability of the 3D measures of wrists and MCPs in subjects with arthritis in a clinically relevant setting. To compare inter-session reliability, 7 wrist and 6 MCP regions from 5 subjects (3 JIA and 2 RA) were scanned twice by an experienced camera operator. The subject left the room between each of the imaging sessions. Wrist and MCP volume and SDI measures demonstrated excellent reliability across imaging sessions (Table 1). In wrists, pooled inter-session volume SD was 0.9 ml (CV, 1.3%) and SDI SD was 0.1 mm (CV, 1.1%). Inter-session pooled MCP volume SD was 0.1 ml (CV, 1.3%) and SDI SD was 0.1 mm (CV, 1.1%). ICCs 13 for all 3D measures of wrists and MCPs were greater than 0.99 (wrist volume ICC = 0.992, wrist SDI ICC = 0.996, MCP volume ICC = 0.995, and MCP SDI ICC = 0.999). Based on the inter-session data, volume changes greater than 1.1 ml in the wrist and 0.5 ml in the MCPs between imaging sessions would be considered significant with 99% confidence. Similarly, a change in the SDI of 0.4 mm in the wrist or 0.3 mm in the MCPs between imaging sessions would also be significant with the same degree of confidence.

We performed a set of experiments to determine the ability of the 3D imager to detect surface changes due to joint swelling, using clay on a mannequin hand to simulate different degrees of swelling consistent with JIA or RA (Figure 2). A known volume of clay was applied to the wrist or MCP region either in a lump, to simulate focal swelling, or spread over a large area, to simulate diffuse swelling (Figure 2a). The estimated changes in volume and SDI were based on the average of three models with the mannequin hand held in fixed position. Three-dimensional imaging proved accurate and sensitive in identifying small changes in both volume and SDI (Figure 2b,c). A significant increase from baseline volume was detectable with the addition of as little as 0.2 ml of clay (0.8% above baseline volume) to the MCP ROI (p = 0.0001) and 0.6 ml (1.5% above baseline volume) to the wrist ROI (p = 0.0002). A significant increase in SDI from baseline due to simulated swelling was also detected with the addition of as little as 1.3 ml of clay (3.2% above baseline volume) to the wrist ROI (p = 0.001) and 0.6 ml (2.5% above baseline volume) added to the MCP ROI (p = 0.002). The SDI was able to discriminate between focal and diffuse swelling when 1.6 ml of clay was added to the wrist ROI (p = 0.02) and 0.6 ml to the MCP ROI (p = 0.0003).

To determine the reliability of thermal imaging of the wrist and MCP, 6 normal adult wrists and hands from 3 controls were imaged on 3 separate days. Three thermal scans were obtained at each session and the HDI was calculated for each ROI. Intra-session (that is, same day and time) HDIs were very similar, with SDs less than 0.05°C (data not shown). Pooled inter-session (that is, day-to-day) SD of wrist HDIs was 0.2°C (Figure 3a) whereas MCP HDI performed less well, with an inter-session SD of 0.4°C (Figure 3b). Pooled inter-session CVs were 22.1% for the wrist and 29.7% for the MCP, indicating relatively large day-to-day variation. This was also reflected in the low HDI ICC 13 values for wrists (0.146) and MCPs (-0.295). However, no control wrist or MCP HDI exceeded 1.3°C, suggesting that an HDI above 1.3°C might be indicative of the presence of arthritis. To explore this further, we compared HDI values of 10 control wrists and 10 control MCPs to 18 wrists with active arthritis and 9 MCPs with active arthritis. As shown in Figure 3c, an HDI cutoff of 1.3°C discriminated well between controls and patients with active arthritis. The mean ± SD HDI in control joints was 1.0°C ± 0.2°C compared with 1.7°C ± 0.6°C in joints with active arthritis (p < 0.0001). By ROC analysis, an HDI value of 1.3°C yielded a sensitivity of 67% and a specificity of 100%. The area under the ROC curve was 0.823. No significant differences in HDI were seen between control adults and children or between arthritic adults and children.

To demonstrate the potential utility of 3D and thermal surface imaging to monitor arthritis, we have longitudinally imaged two patients with wrist arthritis. The first patient was a 9-year-old female with anti-nuclear antigen (ANA)-negative and rheumatoid factor (RF)-negative polyarticular JIA who presented with left wrist pain, warmth, and swelling. The decision was made to proceed with intra-articular steroid injection, and the patient underwent imaging prior to the procedure (Figure 4). The patient returned for re-imaging 5 days after the injection. A reduction in volume of 2 ml, representing a 10% decrease, was noted (Figure 4a). No significant change in SDI was observed, although the area of decreased swelling was evident on the surface color map (Figure 4b). HDI values changed from 1.9°C prior to the injection to 1.1°C after the injection (Figure 4c), associated with narrowing of her temperature frequency distribution (Figure 4d). These quantitative findings correlated with both physician-assessed improvement in swelling and tenderness and patient report of symptom reduction.

The second patient was a 45-year-old Caucasian female with long-standing RF-positive RA on a regimen of hydroxychloroquine and methotrexate. She underwent imaging after completing a course of oral steroids for flare of her disease. At this initial imaging session, her symptoms and physical exam findings were minimal. Ten days later, she returned with complaints of increased swelling, stiffness, pain, and warmth in the right wrist. Her wrist was re-imaged. An increase in swelling, particularly on the dorsolateral aspect of the wrist, was visually apparent in the 3D models. Wrist volume increased between sessions by 4 ml, representing an 8.7% increase from baseline (Figure 5a). Wrist SDI increased between sessions by 1.4 mm, representing an 18.4% increase, along with an obvious change in surface contour as reflected by the surface color map (Figure 5b). The patient's wrist HDI increased from 1.5°C to 2.5°C (Figure 5c). Relative frequency distribution went from narrow to broad (Figure 5d). These quantitative findings correlated with both physician assessment of disease activity and patient report of worsening symptoms.